Reconfigurable Adaptable Micro-robot
R. Lal Tummala1, R. Mukherjee2, D. Aslam1, Ning Xi1, S. Mahadevan3, J. Weng3
1
Department of Electrical and Computer Engineering
2
Department of Mechanical Engineering
3
Department of Computer Science and Engineering
tummala@egr.msu.edu, mukherji@egr.msu.edu, aslam@egr.msu.edu, xin@egr.msu.edu,
mahadeva@cps.msu.edu, weng@cps.msu.edu
Michigan State University, East Lansing, Michigan 48824, USA
ABSTRACT
The research on micro-robots with distributed intelligence,
which are capable of reconfiguring themselves to adapt to
various defense and civilian tasks, is relatively new. A
team of interdisciplinary faculty from the College of
Engineering at Michigan State University has taken this
challenge and are designing micro-robots that incorporate
learning, control and sensing. This paper presents the
goals, and directions of this research along with the details
of a preliminary design of one of the robots.
1.
a)
Size: The robot must measure less than 5cm in any
dimension.
INTRODUCTION
Nishi [1-3] , Backes [4] have built large wall-climbing
robots with suction cups . However, designing a microrobot that can climb walls, walk on ceilings, crawl through
pipes and traverse on floors is a challenging task. An
interdisciplinary faculty in the College of Engineering at
Michigan State University
have been working on
designing such a robot with the support from the Defense
Advanced Research Projects Agency (DARPA). The
major objective of this research is to design , build and
test a prototype micro-robot, that is 1) able to walk,
crawl and flip over to climb walls, stairs and navigate
through narrow spaces and 2) able to work individually or
team up with other robots to perform a mission. We call
these robots reconfigurable, adaptable micro-robots. They
are called reconfigurable, because they are designed to
mechanically reconfigure themselves to provide different
forms of locomotion. (Figs 1-3). They are called
adaptable because they can adapt to different
environments, such as, flat ground, vertical walls, ceilings,
stairs, uneven surfaces etc. The design calls for size
constraint of 5cm in any direction and completely
autonomous. The purpose of this paper is to provide the
status of this effort.
2.
capabilities including autonomous operation and
adaptability to different environments such as walls,
ceilings etc. Adaptability to operate on walls and floors
has been achieved using suction cups. In order to achieve
this mode of operation autonomously, we must consider
the following:
Fig. 1. Walking motion
ROBOT MECHANISM
Fig. 2. Flip over motion
The design of this robot was complicated by the
constraints imposed by its miniature size and desired
b) Weight Limitations: Suction cups that are limited in
size support the robot. Since the holding capacity of
a suction cup is proportional to both achievable
vacuum level and suction cup diameter, the holding
capacity of the foot is also limited. Consequently, net
robot weight is the most substantial constraint.
c)
Actuator Limitations: While the frame structure of
such a robot can be quite lightweight, actuators
contribute significantly to net robot weight. The
number of degrees of freedom, number of actuators
and the robot weight are related. As we increase the
number of degrees of freedom, the robot weight will
increase and will become excessive. For this reason,
special attention must be given to use as few actuators
as possible through novel joint structure, minimal
degrees of freedom, and frame optimization.
the diagram by the symbols , , and . Note that angle 
is indicated on joints 2 and 3 as  and 2, respectively. A
single actuator mounted on link 2 drives these joints via a
belt drive with a 2:1 reduction coupling them. The  DOF
at joint 1 is driven independently of joints 2 and 3 by an
actuator mounted on link 2 and facilitates walking motion.
The  DOF at joint 4 is driven via a differential (not
shown) intersecting Joint 3 and a by a motor mounted on
Link 3. The  DOF permits steering capabilities for the
robot. By reducing the number of actuators in this manner
minimizes weight while provide additional package space
for solid state components and power sources.
Fig 3. Crawling motion
Fig. 4. Micro-robot
d) Control Strategies: The importance of intelligent and
efficient control strategies is all the more important
given the reduced DOF and maneuverability
discussed above. The computational resources and
available sensors ultimately limit the capabilities of
the control strategies.
e)
Intelligence: On a robot of this size, very limited
space is available for memory, sensors, actuator
controllers, and processing units, all of which
contribute to intelligent autonomous operation. The
design must use all available space to incorporate
these traditionally solid state components.
f)
Power supply: Besides actuators, power supplies
such as batteries contribute most substantially to
weight. Dense power sources as well as efficient
actuators are paramount.
Based on the above considerations, a number of kinematic
structures were evaluated and as a preliminary design, a
biped structure with four joints driven by three actuators
was chosen. A diagram of this robot is shown in Fig. 4. It
consists of two legs supported by the suction cups. The
three degrees-of - freedom of the robot are indicated on
3. SMART ROBOT FOOT (SRF) DESIGN
Robotic foot consists of a suction cup, vacuum pump and
an aluminum connector as shown in Fig 5. A
strengthening ring was provided on the cup neck to
prevent bending under parallel load. It weighs
approximately 31 grams and incorporates pressure sensor
Fig. 5. Smart Robotic Foot
housing between the pump and the suction cup. A shape
memory alloy based on Titanium and Nickel micro-valve
is incorporated to control the release time. The views of
robotic foot on the wall and the metal cabinet are shown in
Fig. 6. More details are given in a forthcoming paper. [5]
Fig. 7. Right foot in contact with surface (RFS)
Fig. 6. SRF on the wall and metal cabinet
3.
KINEMATICS
4.1 Coordinate frames assignment
Usually the analyses of biped walking are based on the
assumption that the motion in the sagittal plane is not
affected by the motion in the lateral plane. For this reason,
in most of the previous studies the kinematics and
dynamics of the biped robots are considered separately
and in different planes. However, since the movements of
our walking robot in sagittal and lateral planes are
coupled, this traditional approach is not applicable here.
Because the structure of robot requires that at least one
foot remains contact with the surface all times, the setup
of the coordinate frames are assigned in three-dimensional
space with respect to right-foot supporting (RFS) phase
and left-foot supporting (LFS) phase. The assignment of
coordinate frames is given using D-H convention in RFS
and LFS phases as shown in Figs. 7 and 8.
Fig. 8. Left foot in contact with the surface (LFS)
4.2 Forward kinematics
In RFS phase, the forward kinematics of the robot can be
written as:
n s a
T04  
0 0 0
 nx
p  n y

1   nz

0
For example, in the RFS phase the base coordinate frame
is attached to the right foot which is fixed on the surface.
The left foot can move freely, with the “hand coordinate
frame" attached.
where:
The vectors p, n, s, a represent the position and
orientation of the end-points. Note, in this design the
length of link 2 equals to the length of link 3 (see Fig.4).
a x  T (1,3)  C1C 234
sx
sy
ax
ay
sz
0
az
0
px 
p y 
pz 

1
n x  T (1,1)  C1C 234
s x  T (1,2)  S1
p x  T (1,4)  C1 (a2 C 2  a3C 23 )  d 3 S1
n y  T (2,1)  S 1C 234
s y  T (2,2)  C 1
a y  T (2,3)   S 1 S 234
p y  T (2,4)  S 1 (a 2 C 2  a3C 23 )  d 3C 1
n z  T (3,1)  S 234
s z  T (3,2)  0
a z  T (3,3)  C 234
p z  T (3,4)  a2 S 2  a3 S 23  d1
where  3  k 2 , S, C represent the function sin, cos ,
respectively.
Fig. 9. System Overview
In LFS phase:
n x  T (1,1)  C 4 C123
s x  T (1,2)  C123 S 4
a x  T (1,3)  S123
p x  T (1,4)  a1C1  a2 C12  d 4 S123
n y  T (2,1)  C 4 S 123
s y  T (2,2)   S 123 S 4
a y  T (2,3)  C 123
p y  T (2,4)  a1 S 1  a 2 S 12  d 3C 123
nz  T (3,1)  S 4
s z  T (3,2)  C 4
a z  T (3,3)  0
p z  T (3,4)  d 2
where  2  k 3 . Note that  1 to  4 are defined in the
corresponding coordinate frames in RFS and LFS phase.
A detailed analysis is given in Yue et.al [6].
4.
SYSTEM OVERVIEW
The system overview is given in Fig. 9. At the lowest
level each joint is controlled by a PIC based controller.
These controllers are of PID type. The inputs for each
controller are : position, velocity and acceleration. The
status of each joint is obtained through an encoder. The
robot has infrared sensors for communication and a microcamera for visual input. These robots are designed to
work individually or as a group. They also can work with
other devices and/or robots using different modes of
communication.
We now describe the proposed software architecture of
the robot in more detail. The overall demonstration plan
that we are working toward requires a team (up to three)
of the wall-climbers to work together to build up a
situational awareness of the environment. This capability
requires the robots to navigate around an environment,
learn the structure of the environment, and transmit
information back to the host receiver.
The software architecture is decomposed into the
following components:
Reactive planner: Behavior-based architectures are a wellknown framework for programming mobile robots [7].
We use this approach for the reactive planner. One
example of the type of behavior that will be implemented
on the wall-climber is “homing” towards a designated
target. The target location could be a window (on a wall),
or it could be a specific point on the window. Both
individual and group behaviors are being implemented.
The group behaviors will require inter-robot
communication, which is being tested using radio as well
as infrared technologies. The robots will send each other
short coded messages, which could represent motion
commands, sensor data, or system status (e.g. battery
level).
Task planner: One problem with behavior-based robots is
the lack of flexibility in achieving different goals (e.g.
homing towards a target may require a different set of
behaviors than tracking a wall, or going through a narrow
pipe). The role of the task planner is to select an
appropriate subset of behaviors for the particular goal
(which could be input by the user). The task planner also
uses a sparse representation model of the environment in
directing the movement of the robot. For example, if the
robot’s goal is to go towards a window, the task planner
uses a model of the wall to direct the movement of the
robot away from obstacles. The model of the environment
need not be explicit. For example, the robot could be
trained online by collecting sample images labeled with
the correct behavioral response in those situations (i.e.
supervised learning, as described in [8] and [10]), or it
could be trained by delayed feedback (i.e. reinforcement
learning, as described in [9]).
8.
S. Mahadevan, G. Theocharous, N. Khaleeli, “Rapid
concept learning for mobile robots”, Autonomous
Robots, vol. 5, pp. 239-251, 1998.
Sensory processing: The wall-climber will be equipped
with a variety of low-dimensional sensors (e.g. ambient
light infrared sensors), as well as high-dimensional
sensors (e.g. a micro-camera). Reactive behaviors, such as
avoiding obstacles or homing, will use both types of
sensors. The task planner will learn a model of the
environment by combining low-dimensional and highdimensional sensor representations.
9.
S. Mahadevan and J. Connell, “Automatic
Programming of Behavior-based Robots using
Reinforcement Learning”, Artificial Intelligence, vol.
55, pp. 311-365, 1992.
5.
10. J. Weng and S. Chen, ``Vision-guided navigation
using SHOSLIF,'' Neural Networks, vol. 11, pp.
1511-1529, 1998.
CONCLUDING REMARKS
Acknowledgment
In this paper, preliminary details of a reconfigurable robot
are discussed. It has a biped structure with suction cups at
its extremities and is designed to climb walls, and crawl
on ceilings. More details along with the experimental
results will be presented at the conference.
6.
REFERENCES
1.
Akira Nishi and H. Miyagi, "Simulation and test of a
flight type wall-climbing robot", Proc. World
Automation Congress (WAC'96), pp.569-576, 1996.
2.
A. Nishi, "Development of Wall-Climbing Robots",
Computers and Electrical Engineering, Vol. 22, No.
2, pp.123-149, 1996.
3.
Akira Nishi, "A Biped Walking Robot Capable of
Moving on a Vertical Wall", Mechatronics Vol. 2,
No.5, pp.543-554, 1992.
4.
Paul G. Backes et.al , " The Multifunction Automated
Crawling System (MACS), ", Proceeding of IEEE
International conference on
Robotics and
Automation, Albuquerque, New Mexico, pp. 335340, 1997.
5.
G. Dongi and D. Aslam " Design, Fabrication and
Testing of robotic foot for micro-robots" , to appear
6.
Meng Yue, Mark Minor, , Ning Xi, and Ranjan
Mukherjee, " Kinematics workspace analysis of a
miniature walking robot", to be presented at IROS
conference, Korea, 1999.
7.
R. A. Brooks, “A robust layered control system for a
mobile robot”, IEEE Transactions on Robotics and
Automation, vol. 2, no. 1, pp. 14-23, 1986.
This work is supported by the Defense Advanced
Research Projects Agency, DARPA contract No.
DAANO2-98-C-4025.